allura-org/MS3-24B-Roselily-Creative

Warm
Public
24B
FP8
32768
Hugging Face
Overview

Overview

allura-org/MS3-24B-Roselily-Creative is a 24 billion parameter model built upon ToastyPigeon/ms3-roselily-instruct. Its primary focus is on creative text generation, incorporating extensive data from roleplay and story writing. The model supports a substantial 32768 token context length, making it suitable for longer creative narratives.

Key Capabilities & Features

  • Creative Text Generation: Specifically fine-tuned for roleplay, story writing, and other creative applications.
  • Instruction Formats: Optimized for ChatML and Alpaca instruction formats, with Tekken v7 also supported.
  • Sampler Robustness: Designed to be less sensitive to sampler settings compared to other instruction-based MS3 models, though conservative samplers are still recommended.
  • Chat Template Flexibility: Includes specific token assignments for ChatML to ensure compatibility and proper generation, while maintaining Tekken token integrity for potential merges.

Use Cases

This model is particularly well-suited for:

  • Roleplaying scenarios
  • Generating creative stories and narratives
  • Interactive fiction and adventure modes

Users should be aware that stopping strings like <|im_end|> or </s> might need explicit configuration depending on the tokenizer and chosen chat format.